Page 58 - CyberSecurityEssentialsEnglish

Chapter Two

Talking to AI, the Craft of Prompting

From a passing question to a clear task and a usable professional output
A detailed treatment of prompting, examples, iterative refinement, long context, and
structured outputs
Chapter question: how do we turn an idea in our head into a work specification that a system
can carry out and that we ourselves can evaluate?

Story: 624 Attempts Before the Right Image
In 2022, an artwork titled Théâtre D’opéra Spatial drew wide attention after it won a fine
arts category at a state fair competition. What matters for this chapter is not the legal

argument about authorship that followed, but the description of how the work was
produced. Before the United States Copyright Office, the artist Jason Allen explained that he
had entered text prompts and revisions into the tool Midjourney at least 624 times before

reaching a first version he accepted, and that he then made further changes using other
tools [1].

The Copyright Office ultimately refused to register the image as submitted, on the ground
that the material generated by the tool was not the product of human authorship. That
outcome belongs to a different discussion. For our purposes, the number itself is the lesson.

The episode breaks a common myth: that a striking result usually comes from a single magic
prompt written once. In practice a task often begins as an incomplete idea and improves

through sharper description, examples, choices, comparison between results, and the
deliberate rejection of weak directions. Even when the final output looks simple or elegant,
a long cycle of trial, evaluation, and refinement may lie behind it.

Yet the number of attempts is not by itself a measure of quality. A person can repeat a
request hundreds of times without ever learning why it fails. Professional prompting does

not mean rephrasing at random. It means deciding what we want, measuring the gap
between the requirement and the output, and then changing one influential factor or
moving to a more suitable tool, source, or workflow. This chapter is therefore about the shift

from writing a request to designing and testing a work specification.

Source: U.S. Copyright Office, Review Board Decision on Théâtre D’opéra Spatial [1].

“A professional prompt is not an incantation. It is a working hypothesis that is tested,
whose output is measured, and that is then improved or replaced.”

57